Deep Learning Based Vehicle Seat Belt Detection Algorithm for Driver and Passenger Seat Occupants

With the increase of vehicle popularity and traffic casualty rate, seat belt as a very important protection measure that can effectively protect the safety of passengers, and Chinese traffic law has imposed penalties on vehicle driver and passenger occupants who do not wear seat belt since 2013. In order to eliminate the disadvantages of low efficiency of manual determination and high rate of misjudgment and omission of violation capture photos, and to realize automatic and fast detection of whether the drivers and passengers wear seat belts, this paper proposes a seat belt detection alm through YOLOv5 network and AlexNet network. By locating the windshield through the position relationship of the front windshield of the vehicle, and then obtaining the windshield boundary by the Hough transform, the positions of the driver and the co-driver are divided, and the seat belt is recognized for the driver and the co-driver respectively. Training and testing are conducted on vehicle dataset, front windshield dataset and seat belt dataset, and the results show that the method can effectively identify whether the primary and secondary drivers are wearing seat belts respectively, and the recognition rate is significantly improved compared with other methods.